Lecture 24: Heat Flow COMPSCI/MATH 290-04 Chris Tralie, Duke - PowerPoint PPT Presentation

Lecture 24: Heat Flow COMPSCI/MATH 290-04 Chris Tralie, Duke University 4/12/2016 COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Announcements ⊲ Group Assignment 3 Out: First Deadline Monday 4/18. Final Deadline Wednesday 4/27 (Sakai says 4/26 but that’s wrong...e-mail me solution if go until 4/27) ⊲ Final Project Final Deadline 5/3 5:00 PM COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Table of Contents ◮ Group Assignment 3 Preview ⊲ Scalar Fields / Laplacian Review ⊲ Heat Flow COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Group Assignment 3 Preview COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Table of Contents ⊲ Group Assignment 3 Preview ◮ Scalar Fields / Laplacian Review ⊲ Heat Flow COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Mesh Scalar Fields (Functions) ⊲ A real number for every point on the surface COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Coordinates As Functions COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Laplacian Eigenfunctions (Homer Modes) Lf = λ f 0 1 2 COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Discrete Circle Laplacian Eigenvectors https://github.com/COMPSCI290-S2016/NumpyDemos/blob/master/1DLaplacian.py COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Discrete Circle Laplacian Eigenvectors COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Curvature Vector δ x = Lx , δ y = Ly , δ z = Lz � δ = ( x i , y i , z i ) − ( x j , y j , z j ) j ∈ N ( i ) COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Table of Contents ⊲ Group Assignment 3 Preview ⊲ Scalar Fields / Laplacian Review ◮ Heat Flow COMPSCI/MATH 290-04 Lecture 24: Heat Flow

1D Heat Equation Let f ( x , t ) be the distribution of heat over a 1D bar of uniform material parameterized by x at time t . Then heat flow is governed by ∂ 2 f ( x , t ) = ∂ f ( x , t ) ∂ x 2 ∂ t COMPSCI/MATH 290-04 Lecture 24: Heat Flow

1D Heat Equation Let f ( x , t ) be the distribution of heat over a 1D bar of uniform material parameterized by x at time t . Then heat flow is governed by ∂ 2 f ( x , t ) = ∂ f ( x , t ) ∂ x 2 ∂ t ⊲ The higher the curvature of the heat distribution, the faster it dissipates COMPSCI/MATH 290-04 Lecture 24: Heat Flow

1D Heat Flow Example COMPSCI/MATH 290-04 Lecture 24: Heat Flow

1D Heat Flow Example COMPSCI/MATH 290-04 Lecture 24: Heat Flow

1D Heat Flow Example COMPSCI/MATH 290-04 Lecture 24: Heat Flow

1D Heat Flow Example COMPSCI/MATH 290-04 Lecture 24: Heat Flow

1D Heat Flow Example COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Generalizing To Surfaces ∂ 2 f ( x , t ) = ∂ f ( x , t ) ∂ x 2 ∂ t Let f be the discrete values of a function on a mesh. What is the heat equation? COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Generalizing To Surfaces ∂ 2 f ( x , t ) = ∂ f ( x , t ) ∂ x 2 ∂ t Let f be the discrete values of a function on a mesh. What is the heat equation? Lf = − ∂ f ∂ t COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Generalizing To Surfaces ∂ 2 f ( x , t ) = ∂ f ( x , t ) ∂ x 2 ∂ t Let f be the discrete values of a function on a mesh. What is the heat equation? Lf = − ∂ f ∂ t Video example... COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Solutions To The Heat Equation ∂ 2 f ( x , t ) = ∂ f ( x , t ) ∂ x 2 ∂ t COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Solutions To The Heat Equation ∂ 2 f ( x , t ) = ∂ f ( x , t ) ∂ x 2 ∂ t f ω = cos ( ω x ) e − ω 2 t COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Solutions To The Heat Equation ∂ 2 f ( x , t ) = ∂ f ( x , t ) ∂ x 2 ∂ t f ω = cos ( ω x ) e − ω 2 t Lf = − ∂ f ∂ t COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Solutions To The Heat Equation ∂ 2 f ( x , t ) = ∂ f ( x , t ) ∂ x 2 ∂ t f ω = cos ( ω x ) e − ω 2 t Lf = − ∂ f ∂ t Let φ k be the k th eigenvector of L and λ k be the associated eigenvalue. Then solutions are f k = φ k e − λ t COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Initial Conditions What if initial condition f 0 doesn’t happen to be an eigenvector of L ? COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Initial Conditions What if initial condition f 0 doesn’t happen to be an eigenvector of L ? ⊲ Project f 0 onto eigen basis, sum the solutions of each eigenvector individually � ( f T 0 φ k ) e − λ k t φ k f ( t ) = k COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Initial Conditions What if initial condition f 0 doesn’t happen to be an eigenvector of L ? ⊲ Project f 0 onto eigen basis, sum the solutions of each eigenvector individually � ( f T 0 φ k ) e − λ k t φ k f ( t ) = k Demo... COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Heat Kernel Signature For every point, the fraction of heat that stays at that point after a certain amount of time t = 20 t = 500 What does this look like? COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Heat Kernel Signature For every point, the fraction of heat that stays at that point after a certain amount of time t = 20 t = 500 What does this look like? Multiscale curvature COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Heat Kernel Signature: Computations � ( f T 0 φ k )[ i ] e − λ k t φ k [ i ] f ( t )[ i ] = k COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Heat Kernel Signature: Computations � ( f T 0 φ k )[ i ] e − λ k t φ k [ i ] f ( t )[ i ] = k By definition, starting with a unit amount of heat at every point � 1 a = i � f 0 [ a ] = 0 otherwise COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Heat Kernel Signature: Computations � ( f T 0 φ k )[ i ] e − λ k t φ k [ i ] f ( t )[ i ] = k By definition, starting with a unit amount of heat at every point � 1 a = i � f 0 [ a ] = 0 otherwise � φ k [ i ] e − λ k t φ k [ i ] f ( t )[ i ] = k � e − λ k t φ k [ i ] 2 f ( t )[ i ] = k COMPSCI/MATH 290-04 Lecture 24: Heat Flow

Another Example t = 20 t = 500 COMPSCI/MATH 290-04 Lecture 24: Heat Flow